Akademik Veri Yönetim Sistemi
VII. International Applied Statistics Congress (UYIK-2026), İstanbul, Türkiye, 11 - 13 Mayıs 2026, ss.1, (Özet Bildiri)
The Lomb-Scargle Periodogram (LSP) is a widely used method for dominant frequency estimation in irregularly sampled time series data, with applications spanning astronomy, geophysics, climatology, and engineering. While LSP performs well for stationary or near-stationary signals, its behavior under non-stationary conditions, particularly when the signal frequency varies linearly over time (chirp signals), has received limited attention in the context of spectral analysis applications. This study investigates the frequency estimation bias introduced by the LSP when applied to linearly frequency-modulated signals, with a focus on characterizing the conditions under which the estimated peak frequency deviates from the true mid-window frequency. A simulation framework was developed in Python using synthetically generated chirp signals with controlled parameters: Base frequency, frequency rate of change, and observation window length. Irregular sampling was applied to replicate realistic acquisition scenarios. For each parameter combination, the LSP was computed and the residual between the estimated and true mid-frequency was recorded. Multiple realizations were averaged to suppress stochastic effects from random sampling. The results show that the LSP residual exhibits a systematic and structured pattern across the frequency-rate and window-length parameter space. Beyond a certain combination of frequency rate and window length, the LSP peak frequency undergoes discrete jumps rather than gradual drift. An empirical power-law model was fitted to the failure boundary separating accurate and biased estimation zones. Tests across different base frequencies indicate that the boundary shape remains consistent, indicating a general structural property of the LSP rather than a frequency-dependent artifact. Additional tests under varying signal-to-noise ratios confirm that the observed peak-jump behavior persists regardless of noise level, supporting its deterministic nature. These preliminary results suggest that the failure boundary is predictable and potentially modelable, motivating further analytical investigation.